HLE No ToolsactiveReasoning

HLE No Tools: Humanity's Last Exam — No Tools

Metric: Accuracy (higher is better)Introduced: 2025

Humanity's Last Exam expert-level cross-domain reasoning benchmark measured without external tools. Kept separate from with-tools and Pro-compute variants.

Models ranked

14

tracked on this benchmark

Score band

49.8 – 5.2

best → lowest tracked

Snapshot trend

-7.89

Oct 15 → Oct 19 · 3 models

Leaderboard

Tracked models ranked by Accuracy (higher is better).

Compare candidates
#Model variant and provenanceScore
1
Claude Opus 4.8
Version: HLE no toolsHarness: Not recordedEvaluator: Not recordedObserved: May 28, 2026Confidence: Not recordedSource

Notes: Official Anthropic launch benchmark.

49.8
2
Claude Haiku 5.5
Version: Humanity's Last Exam, no toolsHarness: Not recordedEvaluator: Not recordedObserved: Oct 13, 2026Confidence: Not recordedSource
45.9
3
MiniMax M3.1 Flash

Configuration: MiniMax M3.1 Flash

Version: HLE (w/o tools), text-only 2,158-question subsetHarness: No toolsEvaluator: MiniMaxObserved: Oct 19, 2026Confidence: self-reportedSource

Notes: MiniMax M3.1 Flash launch blog (2026-10-19) main benchmark table image m31-flash-en-2 (webp converted to png; values transcribed by eye from table) incl. footnotes. Footnote: restricted to the 2,158 text-only questions (subset, not full 2,500). Retrieved 2026-10-19.

43.4
4
GPT-5.5 Pro
Version: HLE no tools; standard GPT-5.5 weightsHarness: Not recordedEvaluator: Not recordedObserved: Apr 24, 2026Confidence: Not recordedSource

Notes: No-tools variant kept separate from with-tools and Pro-compute HLE rows.

41.4
5
Beam
Version: HLE no toolsHarness: Not recordedEvaluator: Not recordedObserved: Oct 16, 2026Confidence: Not recordedSource

Notes: Reflection first-party final model card (HF https://huggingface.co/reflection/Beam-1.0-501B-A23B, BF16 column), retrieved 2026-10-16. Replaces 2026-10-05 preview-blog score 36.2. Card label 'HLE (no tools)'. Lab self-reported.

36.0
6
K2 Horizon 375B A23B
Version: Humanity's Last Exam (without tools)Harness: Not recordedEvaluator: Not recordedObserved: Oct 12, 2026Confidence: Not recordedSource
32.0
7
Nemotron 3.5 Bolt 74B-A9B
Version: HLE (no tools)Harness: Not recordedEvaluator: Not recordedObserved: Oct 15, 2026Confidence: Not recordedSource
28.7
8
Nemotron 3 Ultra
Version: HLE no tools (official NVIDIA HF model card)Harness: Not recordedEvaluator: Not recordedObserved: Oct 7, 2026Confidence: Not recordedSource
26.7
9
K2 Horizon 32B
Version: Humanity's Last Exam (without tools)Harness: Not recordedEvaluator: Not recordedObserved: Oct 12, 2026Confidence: Not recordedSource
25.2
10
K2 Horizon 7B
Version: HLE (text), no toolsHarness: Not recordedEvaluator: Not recordedObserved: Oct 12, 2026Confidence: Not recordedSource

Notes: Scores from the released-checkpoint "after" table on the Hugging Face model card.

19.5
11
K2 Horizon 3.7B
Version: HLE (text), no toolsHarness: Not recordedEvaluator: Not recordedObserved: Oct 12, 2026Confidence: Not recordedSource
14.0
12
DiffusionGemma 26B A4B IT

Configuration: DiffusionGemma 26B A4B IT (Entropy-Bounded sampler)

Version: HLE, no toolsHarness: No toolsEvaluator: Google DeepMindObserved: Oct 19, 2026Confidence: self-reportedSource

Notes: Google HF model card evaluation table (instruction-tuned model, recommended Entropy-Bounded sampler; column 'DiffusionGemma 26B A4B'). Row 'HLE no tools'. Retrieved 2026-10-19.

11.0
13
o1-pro

Configuration: o1-pro (API)

Version: HLE (finalized 2,500 Q), no tools; Scale AI SEAL leaderboardHarness: No tools (Scale AI HLE evaluation, o3-mini judge)Evaluator: Scale AIObserved: Oct 19, 2026Confidence: confirmedSource

Notes: Scale AI SEAL Humanity's Last Exam leaderboard (benchmark co-owner, evaluator of record), page JSON entry model 'o1 Pro': score 8.12, CI ±1.07, calibrationError 82, createdAt 2025-04-10. Scale contaminationMessage verbatim: '9% (216 prompts) failed due to a post-training bug and were counted as failures. OpenAI has been informed and is working on a fix. --- Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions.' Score is therefore a lower bound. Methodology per board: o3-mini-2025-01-31 judge, no tools. Epoch hle_external maps the same entry to o1-pro-2025-03-19 (not used directly). API model is deprecated, with retirement on 2026-10-23. Retrieved 2026-10-19.

8.1
14
Gemma 4 12B IT
Version: HLE no tools (official Gemma 4 model card)Harness: Not recordedEvaluator: Not recordedObserved: Oct 7, 2026Confidence: Not recordedSource
5.2

How to read this benchmark

This benchmark scores models where higher is better. Use scores for directional filtering and shortlisting, not universal quality ranking; then validate pricing, context window, provider availability, and fit for your workload.

Trust this score when

  • There is a fresh timestamped snapshot (or multiple snapshots) for this benchmark.
  • The model list covers the same version family you can actually deploy today.
  • Top candidates overlap with your required routing and feature requirements.

Be cautious when

  • There is only one benchmark snapshot or the dataset appears stale.
  • The benchmark metric direction is opposite of your decision objective.
  • The score difference between options is narrow and likely within implementation variance.

Related benchmarks

ANLS*
Average Normalized Levenshtein Similarity*
ARC
AI2 Reasoning Challenge
ARC-AGI-2
ARC-AGI-2 High
ARC-AGI-2 — High Effort
ARC-E
AI2 Reasoning Challenge-Easy
ARC Prize / ARC Challenge

Last reviewed: Jun 17, 2026

Resources

Website